Workflow automation · n8n · Integrations
Routine work that keeps
running in the background.
I connect your applications into automated workflows: data moves from one system to the next without retyping, and an AI step is added only where free text has to be understood.
Discuss your projectFree initial conversation · Directly with Daniel Gläser
- 01Trigger
- 02Data + AI step
- 03Check exceptions
- 04Result in system
Typical workflows
What many offices can hand over to automation.
Good candidates are tasks that happen often, follow the same pattern and are done today by copying and pasting between two programs.
Enquiry form to CRM
Input: an enquiry from the contact form on your website. Result: the contact and a deal are created in the CRM, the enquiry is assigned to the right person and the sender receives a confirmation. Optionally, an AI step summarises the request in two sentences.
Invoice data into accounting
Input: PDF invoices arriving by email. Result: the document is filed, supplier, invoice number and amount are in your accounting software, and mismatches with the purchase order go to the responsible person. How AI reads the documents is covered on the document processing page.
Document processingA report every Monday
Input: figures from your shop, CRM or time tracking. Result: every Monday morning a summary of the previous week is waiting in a Teams channel or the management inbox, with no one stitching spreadsheets together.
Order status in sync
Input: different statuses in the shop, the inventory system and the carrier. Result: statuses are compared regularly and updated in the leading system. Orders stuck for days appear on a follow-up list.
Onboarding new staff
Input: a new entry in the HR list. Result: user account, mailbox and group memberships are prepared, and tasks for hardware and induction are assigned. The workflow does not grant far-reaching permissions on its own; a person approves those.
Rules or AI
Fixed rules first. AI only for unstructured text.
Most workflows need no AI at all. If a form field says “complaint”, a rule is enough to route the enquiry to customer service. Rules are fast, cost next to nothing per run and give the same result for the same input every time.
An AI step earns its place when the workflow has to understand free text: an email in which a customer describes a delivery problem in their own words, an attachment without a fixed layout, or a note from which a name, date and request need to be extracted. The language model returns structured fields, and the rest of the workflow continues on fixed rules.
Because a language model can get things wrong, I build a check around the AI step. If the classification is uncertain or a required field is missing, the case goes to a person instead of moving on automatically. Where a workflow should decide on its own and use tools, it becomes an AI agent with its own rules for permissions and approvals.
Choosing a platform
n8n, Power Automate or custom code: it depends on your environment.
I am not tied to any vendor. The platform follows from your existing systems, the kind of data involved and how much operational work you want to carry yourself.
Self-hosted n8n
n8n is a workflow platform with a visual editor that can run on your own server, for example as a container on Proxmox in your office or with an EU hosting provider. It is published under the Sustainable Use License, a so-called fair-code licence: the source code is openly available and use for your own internal business purposes is permitted. Offering n8n to customers as a service, for example, is not. We check a project against the licence text beforehand; that is no substitute for legal advice.
- Workflows, data and credentials stay on your server.
- Ready-made nodes for many services, HTTP requests for any other API.
- AI steps with cloud models or local models, e.g. via Ollama.
- Single sign-on, Git version control and sharing workflows between users require the paid Enterprise licence.
Microsoft Power Automate
If your business already runs on Microsoft 365, Power Automate is the obvious candidate. Flows using standard connectors such as Outlook, SharePoint or Teams are available with many Microsoft 365 accounts.
- Runs in the Microsoft cloud, no server of your own needed.
- Sign-in and permissions through your existing Microsoft accounts.
- Premium and custom connectors and the on-premises data gateway need an additional licence.
- Licence costs scale with users or individual flows.
Custom integration in code
For high data volumes, complex logic or strict requirements for testing and traceability, I write the integration as a dedicated service, e.g. in TypeScript or C#.
- Version-controlled, automatically tested and traceable in code.
- No platform licence, but more development effort.
- Runs as a container on your server or in an EU cloud.
- A good fit when the integration becomes part of your own application.
Operations
A workflow is only finished once it handles failure properly.
APIs go down, passwords expire and vendors rename fields. A workflow in production is prepared for that.
Failure alerts
If a run fails, a named person is notified straight away by email or Teams, with the affected record and the error message.
Retries
Short API outages are absorbed by a limited number of retries. Where the API allows it, the workflow is built so that a second attempt does not create a record twice.
Logs
Every run can be traced: when it ran, which data it processed and where it stopped. We agree retention periods for logs that contain personal data.
Credentials stored safely
API keys and passwords sit encrypted in the platform's credential store or a password vault, never inside the workflow itself. Each account gets only the permissions the workflow needs.
Documentation
Each workflow has a short description: trigger, systems involved, the person responsible and what to do by hand if it stops.
Maintenance when APIs change
When a vendor changes its API, I adapt the workflow. Platform updates are tested before they go into production.
Server, updates & backup
If n8n runs on your own server, I also look after the infrastructure underneath: the virtual machine or container on Proxmox, backups of the database holding workflows and credentials, and remote access.
IT infrastructureLimits & cost
What automation will not fix, and what it costs.
Automating an unclear process does not make it clearer, only faster. So we first write the process down: who does what, when, and which exceptions occur. Software without an API can often only be connected indirectly, through export files or a mailbox. That works, but it is more fragile.
Cost depends on the number of connected systems, the quality of their APIs, the number of exceptions and the chosen way of running it. Ongoing items include platform licences, server operation, model usage for AI steps and maintenance. We start with one workflow whose effort is agreed before implementation, and see in daily use how much time it saves.
Working together
A clear path to everyday use.
Map the workflow
Record steps, systems, exceptions and frequency together.
Choose the platform
n8n, Power Automate or custom code, matched to your environment and data.
Build and test
Test with real example cases; set up error paths and alerts.
Hand over and maintain
Hand over documentation, monitor runs and adapt when systems change.
Before we begin
Good questions. Clear answers.
Can my business use n8n for free?
The self-hosted Community edition of n8n is published under the Sustainable Use License, which permits use for your own internal business purposes without a licence fee. It does not cover, for example, offering n8n to customers as a service so they can build their own workflows. Some features such as single sign-on require a paid licence. If in doubt, we check the licence text for your case; that is not legal advice.
n8n or Power Automate: which fits better?
If your team works entirely in Microsoft 365 and the workflows stay there, Power Automate is often the shorter route. If data should stay on your own server, many non-Microsoft services need connecting or local AI models are involved, n8n usually fits better. I recommend one after looking at your systems.
Do I need AI for automation at all?
Often not. If the data is already structured, such as form fields or tables, fixed rules are enough. AI makes sense when free text has to be understood, for example in emails or documents without a fixed layout.
What happens when an automated workflow fails?
A named person is notified, short outages are absorbed by limited retries, and the log shows where the run stopped. For every workflow it is documented how to complete the task by hand if needed.
Can n8n work with a local language model?
Yes. n8n can call models running on your own server, for example via Ollama. The text in the AI step then never leaves your network either. Whether a local model handles the task well enough is something we test on your own examples.
What does workflow automation cost?
It depends on the number of systems, the quality of their APIs, the exceptions and how it is run. On top come ongoing costs for licences, servers, model usage and maintenance. We begin with a single workflow and agree its effort before implementation.